Traffic congestion risk assessment method for SRv6 backbone network

By using LSTM neural network in the SRv6 backbone network to predict future traffic demand and calculate the congestion risk coefficient matrix, the network evaluation lag and lack of global problems in the existing technology are solved, and effective congestion risk assessment and path optimization of the backbone network are realized. Improve operator operation and maintenance efficiency and network performance.

CN115604724BActive Publication Date: 2024-11-19JIANGSU FUTURE NETWORKS INNOVATION +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202211270529.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-18
Publication Date
2024-11-19
Estimated Expiration
2042-10-18

AI Technical Summary

Technical Problem

The existing network congestion assessment scheme has lag problems, and it is impossible to effectively evaluate the global congestion risk of the backbone network. The assessment is not global, making it difficult to accurately judge the congestion risk in the backbone network scenario. When a single threshold division cannot effectively evaluate high link load packet loss rate.

Method used

Using the global perspective congestion evaluation method based on SRv6, by obtaining the global network bandwidth matrix and predicting future traffic demands, using the LSTM neural network to predict traffic demands in the future, calculate the congestion risk coefficient matrix of the network path, and set the path congestion switching threshold. , dynamically adjust the path to reduce the risk of congestion in the overall network.

Benefits of technology

It realizes early perception and quantitative evaluation of the congestion level of network paths, improves operator operation and maintenance efficiency, reduces costs, provides global evaluation to guide network strategy optimization, and improves network performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115604724B_ABST
    Figure CN115604724B_ABST
Patent Text Reader

Abstract

The present invention discloses a SRv6 backbone network traffic congestion risk assessment method, which mainly considers two indicators: one is the path bandwidth between the source and destination nodes, which is used to measure the network's carrying capacity for traffic. The other is the predicted network traffic in the future. Since network congestion assessment serves future tuning decisions, it is not enough to evaluate only based on the current business traffic. The recurrent neural network in machine learning is used to predict the demand traffic of the destination node to the source node. The prediction result is used to characterize the average level of traffic bursts in the future.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of congestion risk assessment, and mainly relates to a SRv6 backbone network traffic congestion risk assessment method. Background Art

[0002] In recent years, in order to meet the needs of building 5G networks and distributed large-scale data centers, SRv6 (Segment Routing over IPv6) based on software-defined networking (SDN) has begun to rise rapidly. As a key part of building intelligent IP networks in the 5G and cloud network era, it combines the simplicity and easy expansion of segment routing and IPv6, and has multiple programming spaces. It follows the design concept of SDN and can better meet the current diverse and heterogeneous network business needs and enhance the flexibility of network orchestration and deployment. Therefore, SRv6 technology has been successfully applied in building cross-domain cloud backbone dedicated lines, replacing the traditional MPLS network bearer mode, avoiding problems such as numerous protocols, complex network status, inconvenient business operation and maintenance, and insufficient network reliability. As SRv6 technology is adopted in large-scale deployments, performance monitoring of SRv6 networks is also under study, from data plane monitoring (packet loss and delay) to control plane (sending commands to monitoring entities within nodes), to management plane (collecting metrics), as well as monitoring of network infrastructure status. Corresponding management and monitoring tools are also beginning to be used, continuously and accurately monitoring the health of network infrastructure and the performance it provides, providing a comprehensive view of the entire network for network operation and maintenance. The network dynamic perception provided by these tools will become an important driving force for network routing, congestion control, network health management, and security assurance.

[0003] Then, with the support of existing network monitoring tools and systems, based on the existing infrastructure, how to evaluate the degree of network congestion and estimate the global congestion risk based on segment routing networks is an important basis for network intelligent maintenance and intelligent management. This has guiding significance for improving network operation and maintenance efficiency, estimating network performance status, and corresponding network strategy planning and network traffic scheduling.

[0004] The existing solutions for evaluating the overall network traffic and congestion level mainly consider the current network traffic, that is, judging the network's traffic carrying capacity and whether there is congestion through QoS influencing factors such as the current network link utilization and the remaining bandwidth of the route.

[0005] This type of solution can meet the real-time monitoring of the overall network traffic carrying capacity and congestion level, but the disadvantage is that if this evaluation solution is used and congestion is determined to occur, traffic scheduling will have a certain degree of lag, resulting in a waste of resources and ineffective resource allocation.

[0006] The existing solution for congestion risk assessment of source-destination nodes is based on the support of recurrent neural networks (RNNs). Real-time data from network monitoring is used for traffic estimation and decision-making routing strategies. From the perspective of traffic prediction, traditional RNN neural networks use information obtained at moments close to the current moment, and it is difficult to use information at moments far before the current moment. To solve this problem, an improved RNN, namely the long short-term memory neural network LSTM, was introduced. It improves the hidden layer state update method in the traditional neural network architecture and takes into account the information of both long-correlated and short-correlated time series. It has been proven that it is more suitable for long-correlated dependency problems than traditional RNNs in time series prediction problems, thereby obtaining higher prediction accuracy. Existing technologies have applied it to traffic prediction.

[0007] From the perspective of congestion risk assessment, existing solutions mostly compare the traffic peak between the source and destination nodes within a predicted period of time with the remaining bandwidth of the path to determine whether there is a congestion risk. If not, the path is directly sent to the data forwarding plane for deployment. Otherwise, the congestion risk level of the alternative path is calculated.

[0008] This type of solution estimates the congestion risk and solves the lag problem of traffic scheduling to a certain extent. However, the problem is that on the one hand, the evaluation scheme for evaluating the congestion risk on a specified path is not global, and the use of this evaluation scheme may only achieve local optimization; on the other hand, the rationality of defining the congestion risk threshold is unclear. In some cases, such as the backbone network scenario, a low packet loss rate may also occur when the link load rate is high. Therefore, the congestion risk cannot be effectively evaluated through a single threshold division. Summary of the invention

[0009] The present invention aims to propose a solution for congestion assessment based on the global perspective of the SRv6 backbone network. Congestion assessment mainly considers two indicators. One is the path bandwidth between the source and destination nodes, which is used to measure the network's carrying capacity for traffic. The second is the predicted network traffic in the future. Since network congestion assessment serves future tuning decisions, it is not enough to evaluate only based on the business traffic at the current moment. The recurrent neural network in machine learning is used to predict the demand traffic of the destination node to the source node. The prediction result is used to characterize the average level of traffic bursts in the future.

[0010] In order to achieve the above object, the solution of the present invention is:

[0011] The SRv6 backbone network traffic congestion risk assessment method includes the following steps:

[0012] Step 1, obtain the global network bandwidth matrix;

[0013] Step 2: predict future traffic demand;

[0014] Step 3: Calculate the congestion risk coefficient matrix of the network path and use it as a congestion risk assessment indicator.

[0015] Furthermore, the global network bandwidth matrix in step 1 is expressed as:

[0016]

[0017] In the formula, is the maximum bandwidth of the selected path between source node m and destination node n.

[0018] Further, It is the minimum bottleneck capacity of all links included in the selected path between source node m and destination node n.

[0019] Furthermore, the traffic demand matrix in the future period in step 2 is expressed as:

[0020]

[0021] In the formula, is the traffic demand of destination node n on source node m in the future period.

[0022] Furthermore, the expression of the traffic demand of the destination node to the source node in the future period is:

[0023]

[0024] Among them, f i is the predicted traffic of the destination node to the source node at the i-th moment in the future period, β is the number of moments in the future period,

[0025] Furthermore, for the traffic matrix at any moment in the future, based on the historical traffic matrix of the previous k moments, LSTM is used to predict it, specifically:

[0026] The input of LSTM is the vector form of the traffic matrix, where the traffic matrix F at time t is t The vector form of Where M is the number of source nodes, N is the number of destination nodes, is the traffic from destination node n to source node m at time t;

[0027] The output of LSTM is connected to a fully connected layer;

[0028] The output of the fully connected layer is then processed nonlinearly by the sigmoid layer to output the predicted traffic.

[0029] Furthermore, the congestion risk coefficient matrix of the network path is expressed as:

[0030]

[0031] In the formula, is the congestion risk value of the currently selected path between the source node m and the destination node n for future traffic demand.

[0032] Further,

[0033]

[0034] In the formula, is the global network bandwidth matrix, is the traffic demand matrix in the future period.

[0035] Furthermore, the method further includes setting a path congestion switching threshold. If it is higher than the path congestion switching threshold, it is determined that the currently selected path between the source node m and the destination node n has a congestion risk for future traffic demand.

[0036] A path deployment method, specifically comprising:

[0037] Calculate the congestion risk coefficient matrix of the network path according to the above method in, is the congestion risk value of the currently selected path between the source node m and the destination node n for future traffic demand;

[0038] Calculate the overall network congestion risk in, is the traffic demand ratio of destination node n to source node m;

[0039] Set the path congestion switching threshold. If the path congestion switching threshold is higher than the path congestion switching threshold, the candidate path between the source node m and the destination node n is switched according to the preset priority, and the congestion risk value for future traffic demand after the candidate path between the source node m and the destination node n is switched is calculated;

[0040] Determine whether the following conditions are met after switching the candidate path: Whether the value is not higher than the path congestion switching threshold and the overall network congestion risk C is reduced;

[0041] If there is any candidate path that meets the above conditions, the candidate path will be used for deployment;

[0042] If all candidate paths do not meet the above conditions, the candidate path with the lowest congestion risk C of the entire network is selected for deployment.

[0043] Further,

[0044]

[0045] In the formula, is the traffic demand of destination node n on source node m in the future period.

[0046] Compared with the prior art, the present invention has the following significant advantages: using the technical solution of the present invention, operators can perceive and quantitatively evaluate the congestion level of the path between the source node and the destination node for the upcoming traffic in advance, so as to provide guidance to operators on whether to change the network strategy, and provide a global evaluation of whether the network congestion level has been improved after the operator changes the network. This solution can improve the operator's operation and maintenance efficiency, reduce costs, and increase profits. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is the structure diagram of the LSTM traffic prediction neural network. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below with reference to specific embodiments.

[0049] Compared with traditional networks, SRv6 networks have the following characteristics: when the network boundaries are relatively clear and the source and destination nodes of network services are fixed, path combination and path adjustment are performed at the head node to meet the customized needs of different services and realize a service-driven network. Therefore, when designing a global network congestion assessment solution, the overall SRv6 backbone network congestion assessment based on the source node segment routing strategy is considered. In recent years, the long short-term memory neural network LSTM in the recurrent neural network has a better fitting effect on serialization problems including time series data, and also solves the problem of gradient disappearance in traditional RNN. Applying LSTM to predict demand traffic makes network congestion assessment oriented to future tuning decisions and provides guidance for changing network strategies. The specific implementation plan is as follows.

[0050] 2.2.1 Mathematical Description of Network Model

[0051] We use G = (V, E) to represent the topology of the SRv6 backbone network in the form of a set and a matrix, where V is a set of nodes, where the set of source nodes is M, each source node m∈M, the set of destination nodes is N, each destination node n∈N. The link directly connecting a node i to a node j is denoted by e ij Each link ij Each has a corresponding link capacity (bandwidth) expressed as c ij .

[0052] The future traffic demand of destination node n for source node m is recorded as We use a matrix to represent the demand flow of all destination nodes N to source node M in the future as follows.

[0053]

[0054] For this matrix, each row represents the traffic demand of each destination node for a specified source node. The demand traffic of a destination node to a specified source node is divided by the sum of the demand traffic of all destination nodes to the specified source node to obtain the demand traffic ratio of a destination node to the source node, which is expressed as follows.

[0055]

[0056] Then the demand flow ratio of each destination node to each source node in the future period can also be expressed in the form of a matrix as follows, where

[0057]

[0058] There are multiple candidate paths that meet the requirements between the source and destination nodes. The main path and the protection path are distinguished according to the priority. The path between the source node m and the destination node n is defined as The maximum bandwidth on a path (including a primary path or a protection path) is the minimum bottleneck capacity of all links included in the path, as shown below.

[0059]

[0060] The maximum bandwidth of the selected path between the source node and the destination node (the maximum bandwidth of the primary path by default, and the maximum bandwidth of the selected protection path in case of congestion, failure, etc.) is expressed as It is expressed as follows.

[0061]

[0062] 2.2.2 Forecast of future traffic demand

[0063] In 2.2.1, the future traffic demand of destination node n for source node m is defined as Then the demand flow of all destination nodes N to source node M in the future period of time is expressed by the matrix as follows.

[0064] (I) Traffic matrix prediction of source and destination nodes at a certain time in the future

[0065] For a certain time t, the flow matrix is ​​simplified as F t (t∈[1,T]), where T is the total number of time slots. The traffic forecasting problem can be defined as a set of historical demand traffic data (F t-k ,F t-k+1 ,…,F t-1 )(k∈[1,t]) This demand predicts the current traffic demand

[0066] Based on the improved recurrent neural network - long short-term memory neural network LSTM, the method of directly predicting the traffic matrix as a whole is adopted, that is, the M rows of the traffic matrix are connected in series from top to bottom, and the matrix F t Convert to vector X t The whole is used as the input of LSTMt moment, and the input dimension is M x N.

[0067] like Figure 1 The figure shows the structure of the LSTM traffic prediction neural network, which includes LSTM layer, fully connected layer and Sigmoid layer. The input data of the LSTM layer is a three-dimensional vector (batch, time_step, input_size), batch is the number of samples required to complete a parameter update training, time_step is the set sliding window length, and input_size is the vector dimension of each input data point. The output after the LSTM layer is (batch, time_step, cell_size), where cell_size is the number of hidden units in the LSTM neural layer. The purpose of the fully connected layer after the LSTM layer is to complete the projection of the vector cell_size output of the LSTM layer into a prediction vector of length output_size, and finally pass through the sigmoid layer for nonlinear processing, without changing the data dimension, and output the predicted traffic.

[0068] (II) Predicting the average level of traffic bursts in the future

[0069] As mentioned above, the purpose of future traffic prediction is to assess the risk of network congestion. Network congestion caused by short bursts of traffic does not necessarily cause packet loss, but packet loss is caused by bursts of more traffic over a longer period of time. Therefore, from the above prediction of traffic at a certain moment in the future to the prediction of traffic sequence, a statistically predicted traffic peak is obtained to characterize the average level of traffic bursts in the future.

[0070] We select a predicted traffic sequence [f1,f2,…,f β ], the peak value of the predicted flow is the superposition of the mean and standard deviation of the flow sequence, and the peak value of the predicted flow is denoted as f peak , the expression is as follows.

[0071]

[0072] Among them, μ is the mean of the predicted sequence, and the expression is as follows:

[0073]

[0074] For each pair of traffic demand from destination node n to source node m, the predicted traffic peak is obtained by the above formula, that is, It is expressed as a matrix It is the set of average burst levels of traffic demanded by all destination nodes N to source node M in the future.

[0075]

[0076] 2.2.3 Network congestion assessment

[0077] In summary, we get the global network bandwidth matrix Demand flow matrix for a period of time in the future The congestion risk coefficient matrix of the network path is set as Through the matrix Dot Division Matrix The expression is obtained as follows.

[0078]

[0079] Congestion coefficient matrix Each element in As shown below, it is used to indicate the congestion risk of the current path between the specified source node m and the destination node n for the upcoming traffic.

[0080]

[0081] The definition in 2.2.1 As a weight measure of the importance of the destination node to the source node, that is, if the destination node has a large flow demand for the source node, the impact of the path congestion risk will be greater. and Combined, we get the overall network congestion risk C, expressed as follows.

[0082]

[0083] Set the path congestion switching threshold between the specified source node and the destination node When there is a congestion risk coefficient in the congestion coefficient matrix that is higher than the corresponding path congestion switching threshold, that is, The candidate paths are switched in order of priority, and the congestion risk coefficient of the path is calculated to see whether it meets the requirements and whether the overall congestion risk C of the network is reduced. If it meets the requirements, the path is deployed. If all candidate paths do not meet the requirements, the path with the lowest overall network congestion risk is selected for deployment.

[0084] The present invention proposes a solution based on the global perspective congestion assessment of the SRv6 backbone network, which aims to perceive the network congestion risk in advance and effectively guide the decision-making and deployment of the path. The key technical point of this solution is to combine the traffic prediction technology based on machine learning with the traditional congestion assessment. It mainly considers two indicators. One is the path bandwidth between the source and destination nodes, which is used to measure the network's carrying capacity for traffic. The second is the predicted network traffic in the future period, which is used to characterize the average level of traffic bursts in the future period. The global congestion coefficient matrix is ​​obtained by combining the path bandwidth and the future traffic, which is used to evaluate the congestion level of the entire network.

[0085] By using the solution of congestion assessment based on the global perspective of SRv6 backbone network proposed by the present invention, operators can perceive and quantitatively assess the congestion level of the path between the source node and the destination node for the upcoming traffic in advance, so as to provide guidance to operators on whether they need to change the network strategy, and provide a global assessment of whether the network congestion level has been improved after the operators change the network. This solution can improve the operation and maintenance efficiency of operators, reduce costs and increase profits.

[0086] The scheme designed by the present invention for evaluating the congestion level of the entire network by combining path bandwidth and future traffic has the characteristics of future-oriented decision-making and a global perspective, so as to provide guidance to operators on whether they need to change network strategies, and provide a global evaluation of whether the network congestion level and carrying performance are improved after the operators change the network.

[0087] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. The SRv6 backbone network traffic congestion risk assessment method is characterized by: The following steps are involved: Step 1, obtain the global network bandwidth matrix; Step 2: predict future traffic demand; Step 3, calculate the congestion risk coefficient matrix of the network path and use it as a congestion risk assessment indicator; The global network bandwidth matrix in step 1 is expressed as: Where b n m is the maximum bandwidth of the selected path between source node m and destination node n; The traffic demand matrix in the future period in step 2 is expressed as: In the formula, f n m is the traffic demand of destination node n to source node m in the future period; The expression of the traffic demand of the destination node to the source node in the future period is: Among them, f i is the predicted traffic of the destination node to the source node at the i-th moment in the future period, β is the number of moments in the future period, ; For the traffic matrix at any moment in the future, based on the historical traffic matrix at the previous moment, LSTM is used to predict it, specifically: The input of LSTM is the vector form of the traffic matrix, where the traffic matrix F at time t is t The vector form of , where M is the number of source nodes, N is the number of destination nodes, and X m*n t is the traffic from destination node n to source node m at time t; The output of LSTM is connected to a fully connected layer; The output of the fully connected layer is then processed nonlinearly by the sigmoid layer to output the predicted traffic; The congestion risk coefficient matrix of the network path is expressed as: In the formula, r n m is the congestion risk value of the currently selected path between the source node m and the destination node n for future traffic demand.

2. The SRv6 backbone network traffic congestion risk assessment method according to claim 1, characterized in that: The method also includes setting a path congestion switching threshold. n m If the value is higher than the path congestion switching threshold, it is determined that the currently selected path between the source node and the destination node has a congestion risk for future traffic demand.

3. A path deployment method, characterized in that: Specifically include: Calculate the congestion risk coefficient matrix of the network path according to the method as claimed in claim 1; Calculate the overall network congestion risk , where q n m is the traffic demand ratio of destination node n to source node m; Set the path congestion switching threshold. If r n m If the path congestion switching threshold is higher than the path congestion switching threshold, the candidate path between the source node m and the destination node n is switched according to the preset priority, and the congestion risk value for future traffic demand after the candidate path between the source node m and the destination node n is switched is calculated; Determine whether the following conditions are met after switching the candidate path: n m Whether the value is not higher than the path congestion switching threshold and the overall network congestion risk C is reduced; If there is any candidate path that meets the above conditions, the candidate path will be used for deployment; If all candidate paths do not meet the above conditions, the candidate path with the lowest congestion risk C of the entire network is selected for deployment.

4. A path deployment method according to claim 3, characterized in that: In the formula, f n m is the traffic demand of destination node n on source node m in the future period.

Citation Information

Cited By

  • Financial intelligent backbone network flow scheduling method and system based on SRv6

    CN120915738A

  • SRv6-based financial intelligent backbone network traffic scheduling method and system

    CN120915738B